Collection costs add up.
Dedicated hardware, operator time and resets make it costly to repeat demonstrations across tasks and locations.
Teach robots
the real world.
Targeted robotics training data,
collected through guided XR missions.
+01THE CHALLENGE
Robot training still has
blind spots.
The examples you need are missing.
Existing datasets may not cover your exact task, objects or success criteria. More footage does not guarantee the right examples.
Different instructions. Different starting points.
Recordings of the same task may use different instructions, object positions and starting conditions. Without that context, they are harder to combine into a consistent training set.
More recordings. Uneven quality.
Different demonstrators bring different levels of skill. Mixing weak and strong demonstrations can make learning harder.
+02THE SOLUTION
The experience you’re missing.
Collected through XR.
01
Define once. Deploy at scale.
Define your task and environment. Designed to deploy each mission automatically to tens or hundreds of thousands of existing XR headsets.
02
Your exact task.
Specify the missing skill, objects and success criteria. Give every contributor clear, step-by-step XR instructions.
03
Your controlled environment.
Define the scene layout, object positions and permitted variations. Capture coordinated head, upper-body and hand motion on supported setups.
04
Many people. Shared criteria.
Designed to collect from many contributors under one acceptance protocol, reviewing task completion and recording quality before delivery.
05
Robot-specific motion mapping.
Automate retargeting for selected robots, adapting movements to their joints and physical limits with robot-specific validation.
+03THE NETWORK
Connecting robot challenges
to real human experience.
Meet the A2ZRO platform.
Robotics teams
A2ZRO Missions
Performed by XR contributors
01
Specify what your robot needs to learn
Pick up
02
Users receive guided XR missions
Task completed
03
Full body movement is tracked
04
Bring targeted data into model training
Submit taskXR missionData extractionAccepted data
A2ZRO
Submit taskXR mission
Data extractionA2ZROAccepted data
+04A2ZRO MISSIONS
Real tasks.
Guided by XR.
Follow a task from human movement to robot replay. Explore the motion and attention behind each action.
Become a Contributor+05YOUR FIRST PILOT
Start with
one skill.
Bring a task your robot struggles with. We’ll scope a focused data collection pilot together.
Discuss your pilotBring a concrete task.
Tell us what your robot needs to do, where it falls short and which examples are missing.
Agree on a useful dataset.
Together, we define the variations, tracking signals, demonstration count and acceptance criteria.
Plan how you’ll evaluate it.
Agree on the delivery format and how your team will test the data in its training pipeline.
+06THE FOUNDERS
Meet the founders.
Robin Gansäuer
- Master’s in Computer Science & Business
- AI research at the University of Oxford
- Data science background
- VC experience at Earlybird & Oxford Seed Fund
Tobias Reidl
- Computer science dropout
- AR developer
- AI Product Manager at SAP
- Product Manager at an XR Med-Tech startup
Good questions.
Clear answers.
A pilot would agree the permitted first-person recordings, task annotations, mission metadata, format and acceptance criteria before collection. Precise robot joint actions, contact forces, tactile data and universal motion transfer are outside the initial scope.
The initial prototype is planned around Meta Quest 3 and Quest 3S. Device support, camera access, recording permissions and commercial data-use rights remain subject to technical and platform-policy validation.
We are gathering expressions of interest for a future contributor cohort. Missions are not yet publicly available. Participation would be for adults with eligible hardware, with agreed rewards for accepted contributions where offered.
The proposed collection workflow uses informed consent and task-specific recording. It should minimize private-home details, bystanders and unnecessary personal information. SDK hand-tracking exports are not assumed to be available for sale or external training use.
Give intelligence
some experience.
Bring a collection challenge.
Or join our future XR network.
Let’s scope your data pilot.
What does your model need to learn?